Projects

### Overview: This repository contains PyTorch implementations of various Bayesian neural network inference methods, including Bayes by Backprop (BBP), Local Reparametrization Trick, MC Dropout, SGLD, Kronecker-Factorised Laplace Approximation, Stochastic Gradient Hamiltonian Monte Carlo (SG-HMC), and more. It also includes code for Bootstrap MAP Ensemble. ### Key features: - PyTorch implementations - Support for PyTorch 1.0.1 - Support for both CPU and GPU - Includes code for Bayesian inference methods such as BBP, LR, MC dropout, SGLD, Kronecker-Factorised Laplace Approximation, and SG-HMC - Provides code for Bootstrap MAP Ensemble - Written in Python 2.7 and PyTorch 1.0.1 - Supports both CPU and GPU - Includes Google Colab notebooks for running on a GPU for free ### Tech stack: PyTorch ### Why it matters: This repository is essential for researchers and developers looking to implement and experiment with Bayesian neural network inference methods using PyTorch. It provides a comprehensive set of tools for training and evaluating models using various inference techniques, making it a valuable resource for anyone working in the field of Bayesian neural networks.

Overview: 本项目基于Meta最新发布的Llama-3模型开发，提供了中文Llama-3基座模型和指令精调大模型，旨在提升中文基础语义和指令理解能力。 Key features: - 开源Llama-3-Chinese基座模型和指令模型 - 提供预训练脚本和指令精调脚本 - 提供alpaca_zh_51k, stem_zh_instruction, ruozhiba_gpt4 (4o/4T) 指令精调数据 - 提供快速本地量化和部署教程 Tech stack: Python Why it matters: 本项目是Chinese-LLaMA-Alpaca系列项目的第三期，提供了中文Llama-3基座模型和指令模型，旨在提升中文基础语义和指令理解能力，是研究和应用中文语言模型的重要资源。

### Overview: This project provides a curated list of papers on dataset distillation and related applications, offering comprehensive information on the field. ### Key features: - **Dataset Distillation Papers**: A collection of papers on dataset distillation, including seminal works and recent advancements. - **Contributions Welcome**: Encourages contributions to improve the list of papers. - **Stars and Forks**: 1,978 stars and forks, indicating high community interest and active participation. - **Project Maintainers**: Managed by Guang Li, Bo Zhao, and Tongzhou Wang. ### Tech Stack: HTML ### Why it Matters: This project serves as a comprehensive resource for researchers and practitioners interested in dataset distillation, providing a wealth of information on the field's latest developments and applications.

Overview: $\Phi$-SO is a symbolic optimization package designed for physics, built using Python. Key features: - $\Phi$-SO's symbolic regression module uses deep reinforcement learning to infer physical laws that fit data points. - It supports physical units constraints and class constraints for multi-dataset symbolic regression. - It leverages popular SR packages like SRBench for benchmarking. Tech stack: Python Why it matters: $\Phi$-SO is a powerful tool for physicists and engineers looking to discover physical laws from data, making it easier to model and predict physical phenomena. Repository: WassimTenachi/PhySO GitHub stars: 1978 Source code: [WassimTenachi/PhySO](https://github.com/WassimTenachi/PhySO) Documentation: [physo.readthedocs.io](https://physo.readthedocs.io/en/latest/)

### Overview: Custom Diffusion is a method that allows fine-tuning of text-to-image diffusion models like Stable Diffusion, enabling the customization of new concepts with minimal training data. This technique reduces the storage requirements and inference time by only fine-tuning a subset of model parameters. ### Key features: - **Custom Diffusion with SDXL**: Supports the updated Diffusers code with SDXL. - **CustomConcept101 Dataset**: Released with 101 concepts and their evaluation prompts. - **Fast Training**: Fine-tunes only a subset of model parameters, significantly reducing inference time and storage requirements. - **Multi-Concept Customization**: Allows combining multiple concepts such as new object + new artistic style, multiple new objects, and new object + new category. ### Tech stack: Python ### Why it matters: Custom Diffusion enables the customization of new concepts in text-to-image diffusion models, making it easier to generate images of new objects or styles without extensive training data. This method reduces storage requirements and inference time, making it a valuable tool for various applications in the field of image generation and customization.

### Overview: DeepVoice3_pytorch is a PyTorch implementation of convolutional neural networks-based text-to-speech synthesis models. It includes models for multi-speaker and single speaker text-to-speech, preprocessor for various datasets, and pre-trained models compatible with LJSpeech, JSUT, and VCTK datasets. ### Key Features: - **Convolutional Sequence-to-Sequence Model with Attention**: A model designed for text-to-speech synthesis. - **Multi-speaker and Single Speaker Versions**: Supports both multi-speaker and single speaker text-to-speech. - **Audio Samples and Pre-trained Models**: Available audio samples and pre-trained models for LJSpeech, JSUT, and VCTK datasets. - **Language-Dependent Frontend Text Processor**: Compatible with English and Japanese language processing. ### Tech Stack: PyTorch ### Why It Matters: This project is significant for developers looking to implement text-to-speech synthesis models, particularly for applications requiring high-quality audio output. The pre-trained models and pre-processing tools make it accessible for various datasets, making it a valuable resource for researchers and developers in the field of speech synthesis. ### Repository: r9y9/deepvoice3_pytorch ### GitHub Stars: 1,977 ### Primary Language: Python ### GitHub Description: PyTorch implementation of convolutional neural networks-based text-to-speech synthesis models. Includes models for multi-speaker and single speaker text-to-speech, preprocessor for various datasets, and pre-trained models compatible with LJSpeech, JSUT, and

Overview: This repository is a collection of Python implementations of various Reinforcement Learning (RL) algorithms, focusing on educational purposes. It aims to provide a clear and intuitive understanding of RL concepts without the complexity of advanced libraries. Key features: - Simple Exploration Bot - Q-Learning - SARSA - Expected SARSA Tech stack: Python, Jupyter Notebook Why it matters: This repository serves as an educational resource for beginners in RL, focusing on core concepts without the complexity of advanced libraries. It includes a comprehensive cheat sheet for quick reference and interactive learning through Jupyter Notebooks. Repository: FareedKhan-dev/all-rl-algorithms Primary language: Jupyter Notebook GitHub stars: 1976 GitHub description: Implementation of all RL algorithms in a simpler way GitHub description: This repository is a collection of Python implementations of various Reinforcement Learning (RL) algorithms. The primary goal is to get a deep and intuitive understanding of how these algorithms work under the hood. Due to the recent explosion in the AI domain especially Large Language Models, and many more applications it is important to understand core reinforcement learning algorithms. This repository also includes a comprehensive cheat sheet summarizing key concepts and algorithms for quick reference. This is not a performance-optimized library! I prioritize readability and clarity over speed and advanced features. Think of it as your interactive textbook for RL. The repository currently includes implementations of the following RL algorithms, with more planned: Algorithm Quick Reference: | # | Algorithm | Type | Description | Notebook | |----|-----------|------|-------------|----------

### Overview: The Semantic Segmentation Editor is a web-based labeling tool designed for creating AI training datasets. It supports both 2D and 3D image and point cloud data, making it ideal for autonomous driving research. The tool is developed using Meteor, React, Paper.js, and three.js, providing a robust and efficient labeling experience. ### Key Features: - **Bitmap Image Editor**: Allows users to label images in 2D, supporting JPG and PNG formats. - **Point Cloud Editor**: Enables point cloud labeling in 3D, suitable for autonomous driving applications. - **Docker Support**: Provides a Docker image for easy deployment, improving performance and security. ### Tech Stack: - **Meteor**: The front-end framework used for building the web application. - **React**: The JavaScript library for building user interfaces. - **Paper.js**: A JavaScript library for 2D and 3D graphics and animations. - **three.js**: A JavaScript library for 3D graphics and animations. ### Why It Matters: The Semantic Segmentation Editor is crucial for researchers and developers working on autonomous driving projects. It offers a comprehensive solution for creating high-quality training data sets, which are essential for developing advanced AI models. The ability to handle both 2D and 3D data types makes it highly versatile and applicable to various autonomous driving research scenarios.

### Overview: Nerfies: Deformable Neural Radiance Fields is a project that uses JAX to implement Deformable Neural Radiance Fields (Nerfies), a technique for generating realistic videos from captured images. This codebase is based on JaxNeRF and is designed for efficient training and rendering of high-resolution videos. ### Key features: - Deformable Neural Radiance Fields (Nerfies) - Realistic video generation from captured images - Uses JAX for efficient computation - Supports training and rendering on Google Colab ### Tech stack: JAX ### Why it matters: This project aims to provide a practical tool for generating realistic videos from captured images, which has applications in various fields such as virtual reality, augmented reality, and computer graphics.

Overview: XGrammar is an open-source library designed for efficient, flexible, and portable structured generation, leveraging constrained decoding for 100% structural correctness and supporting various structures like JSON, regex, and custom context-free grammars. Key features: - **Structural Correctness**: Ensures 100% structural correctness of the output. - **Flexibility**: Supports general context-free grammar to generate a wide range of structures. - **Speed**: Achieves near-zero overhead in JSON generation, making it one of the fastest structured generation engines available. - **Universal Deployment**: Supports multiple platforms, hardware, languages, and models. Tech stack: Python Why it matters: XGrammar is highly efficient, flexible, and portable, making it an essential tool for developers looking to integrate structured generation into their projects. Its ability to handle complex structures and support multiple languages and models makes it a versatile solution for various applications.

### Overview: FoundationVision/LlamaGen is a repository that contains pre-trained model weights and training/sampling PyTorch codes for image generation using an autoregressive model approach. This model, LlamaGen, aims to address the limitations of vanilla autoregressive models by applying the next-token prediction paradigm to visual generation. ### Key features: - Two image tokenizers with downsample ratios of 16 and 8. - Seven class-conditional generation models ranging from 100M to 3B parameters. - Two text-conditional generation models with 700M parameters. - Online demos available in [Hugging Face Spaces](https://huggingface.co/spaces/FoundationVision/LlamaGen) for running pre-trained models. - Support for vLLM serving framework for a 300% to 400% speedup. ### Tech stack: PyTorch (>= 2.1.0) ### Why it matters: This repository introduces a new family of image generation models that apply the next-token prediction paradigm of large language models to the visual generation domain. It challenges the notion that vanilla autoregressive models, such as Llama, can achieve state-of-the-art image generation performance if scaled properly by reexamining design spaces, scalability properties, and training data quality.

Overview: This project is an advanced implementation of stacked neural networks to predict stock return predictions. The workflow closely follows the approach outlined in Marcos Lopez de Prado's book "Advances in Financial Machine Learning". The data used for this project is not included in the repository due to size constraints, but the raw data was sourced from Tick Data LLC. Key features: - Multilayer neural network architecture - LSTM regression model for predicting stock returns - Random Forest classification model for predicting stock direction Tech stack: Python Why it matters: This project serves as an educational resource for understanding the principles behind advanced machine learning models used in financial data analysis. It provides a comprehensive guide on how to build and test a neural network model for predicting stock returns, including feature engineering and model training.

### Overview: Opacus is a library that enables training PyTorch models with differential privacy, making it accessible for both ML practitioners and researchers. It supports minimal code changes required on the client, has negligible impact on training performance, and allows clients to track privacy budgets online. ### Key features: - **Privacy Engine**: A tool for making models private using differential privacy. - **Differential Privacy**: Ensures that the privacy of individual data points is protected during training. - **Minimal Code Changes**: Requires minimal modifications to the client's code to enable differential privacy. - **Memory Efficiency**: Significantly reduces memory requirements compared to other privacy-preserving methods. - **Online Budget Tracking**: Clients can track the privacy budget expended in real-time. ### Tech stack: Python ### Why it matters: Opacus simplifies the process of training models with differential privacy, making it easier for practitioners to experiment and learn without extensive knowledge of privacy-preserving techniques. It allows researchers to focus on the core of their research without being overwhelmed by the complexities of privacy-preserving methods.

### Overview: MASR 是一个基于端到端深度神经网络的中文普通话语音识别项目，使用门控卷积神经网络（Gated Convolutional Network）进行训练。通过使用 GLU 激活函数，模型收敛速度更快。项目使用 AISHELL-1 数据集训练，覆盖 4000 多个汉字，工业界通常需要 10 倍以上的录音数据来训练语音识别系统，因此本项目的表现难以与工业界媲美。 ### Key features: - 使用门控卷积神经网络进行语音识别 - AISHELL-1 数据集训练 - 100 个 epoch 的预训练模型 - 11% 的验证集 CER ### Tech stack: Python ### Why it matters: 尽管本项目使用较少的录音数据进行训练，但通过使用门控卷积网络和 GLU 激活函数，模型的收敛速度更快，因此在特定场景下仍能取得较好的识别效果。这对于语音识别项目来说是一个重要的突破。

Overview: This repository is a comprehensive toolkit for developers looking to integrate AI-powered products and businesses with Stripe. It includes SDKs for integrating Stripe's billing infrastructure with various LLMs and agent frameworks, as well as a collection of skills for building autonomous agents with the Model Context Protocol (MCP). Key features: - `@stripe/ai-sdk` for integrating Stripe's billing infrastructure with Vercel's `ai` and `@ai-sdk` libraries. - `@stripe/token-meter` for integrating Stripe's billing infrastructure with native SDKs from OpenAI, Anthropic, and Google Gemini. - Model Context Protocol (MCP) for secure access to the Model Context Protocol server. Tech stack: TypeScript. Why it matters: This repository provides a one-stop solution for developers looking to integrate Stripe with AI-powered products and businesses, making it a valuable resource for those building AI-driven applications.

### Overview: MachineLearningStocks is an educational project designed to introduce you to the workflow of using machine learning for stock predictions. The project uses Python and the scikit-learn library to clean and prepare historical stock price and fundamental data, then applies a simple classifier to predict stock price changes based on fundamental metrics. ### Key Features: - **Data Cleaning and Preparation**: Using `pandas` to preprocess historical stock price and fundamental data. - **Model Training**: Applying a `scikit-learn` classifier to predict stock price changes. - **Backtesting**: Conducting a simple backtest to evaluate the model's performance. - **Prediction Generation**: Generating predictions on current data based on the trained model. ### Tech Stack: Python ### Why it Matters: This project serves as a starting point for understanding the workflow of using machine learning for stock predictions, appreciating the subtleties of the process, and potentially leading to profitable trading systems. While it does not suggest how to combine stocks into a portfolio, it can be a useful starting point for improving risk-adjusted returns through portfolio optimization. ### Repository: robertmartin8/MachineLearningStocks ### GitHub Stars: 1966 ### Data Sources: - **Historical Stock Fundamentals**: Provided by the project. - **Historical Price Data**: Provided by the project. ### Contents: - **Contents** - **Overview** - **EDIT as of 24/5/18** - **EDIT as of October 2019** -

Overview: 《码农周刊》是每周五发送的IT技术周刊，旨在为程序员和编程爱好者提供专业、简单、有用的内容。 Key features: - 每周五发送 - 专业、简单、有用 - 读者圈子专属邮件周报 - 52期专属邮件周报 - 52本好书共读 - 600+圈子成员 - 1000+优质主题 - 独家资料 Tech stack: Python, Markdown, Git, HTML Why it matters: 《码农周刊》不仅提供技术干货，还通过读者圈子的交流和优质主题分享，帮助程序员提升认知和技能，是程序员和编程爱好者不可或缺的资源。

### Overview: OneDiff is an out-of-the-box acceleration library designed to speed up popular diffusion models like HF Diffusers and ComfyUI. It provides optimized GPU kernels and pre-compiled tools for PyTorch, making it easier to integrate into existing projects. ### Key Features: - **Out-of-the-Box Acceleration**: Provides acceleration for popular UIs/libs such as HF Diffusers and ComfyUI. - **PyTorch Compilation Tools**: Compiles PyTorch code into optimized GPU kernels for diffusion models. - **Speedup by 1.7x**: Demonstrated significant performance improvements for Kolors, SD3, and PixArt models. - **3x Faster**: Accelerated SDXL by 3x faster with DeepCache and OneDiff. ### Tech Stack: PyTorch ### Why It Matters: OneDiff simplifies the process of integrating diffusion models into existing projects by providing optimized GPU kernels and pre-compiled tools, thus significantly reducing the time and effort required for model acceleration. This makes it an essential tool for developers working with diffusion models.

### Overview: Nimbalyst is an open-source visual workspace designed for developers using Claude Code, Codex, and OpenCode. It allows you to run multiple coding agents in parallel, edit their work visually, and track tasks and sessions in a unified interface. ### Key Features: - **Visual Editors:** Built-in WYSIWYG editors for markdown, mockups, Mermaid, Excalidraw, CSV, data models, and code in Monaco. - **Red/Green Diff Review:** Step through agent's proposed edits in the rendered document and accept or reject each one. - **Parallel Agent Sessions:** Run several coding agents at once, each isolated in its own git worktree. - **Session Kanban:** Track every session on a board. Search, resume, and link sessions to the files they touched. - **Task Tracking:** Plans, bugs, features, and todos in trackers your agents read and update alongside you. - **Heterogeneous Agents:** Codex and Claude Code side by side, plus OpenCode and GitHub Copilot in alpha and a Gemini provider. - **Git Built in:** Branch and worktree management, staging, AI-drafted commits, and an embedded terminal. - **MCP Client and Servers:** Connect any MCP server, with tool results rendered as visual widgets rather than raw JSON. - **Extension SDK:** Every editor, including the built-in ones, goes through the same `EditorHost` contract, so an editor you build is first-class. Ask your agent to build one for a file type we do not ship. - **Plain Files on

### Overview: This project uses deep learning and natural language processing (NLP) to analyze a large corpus of clear text passwords. The goal is to train a generative model to understand how people change their passwords over time, providing insights into password security and evolution. ### Key Features: - **Password Analysis**: The project analyzes a large dataset of clear text passwords. - **Generative Model Training**: A deep learning model is trained to generate passwords similar to the ones found in the dataset. - **Edit Distance Analysis**: The project calculates edit distances between passwords to identify patterns and similarities. ### Tech Stack: Python ### Why it Matters: This project is significant for researchers and security professionals looking to understand password evolution and improve password security measures. The insights gained can be valuable for developing more effective password policies and security protocols.